BMDExpress Data Viewer - a visualization tool to analyze BMDExpress datasets

Abstract

Regulatory agencies increasingly apply benchmark dose (BMD) modeling to determine points of departure for risk assessment. BMDExpress applies BMD modeling to transcriptomic datasets to identify transcriptional BMDs. However, graphing and analytical capabilities within BMDExpress are limited, and the analysis of output files is challenging. We developed a web-based application, BMDExpress Data Viewer (http://apps.sciome.com:8082/BMDX_Viewer/), for visualizing and graphing BMDExpress output files. The application consists of "Summary Visualization" and "Dataset Exploratory" tools. Through analysis of transcriptomic datasets of the toxicants furan and 4,4'-methylenebis(N,N-dimethyl)benzenamine, we demonstrate that the "Summary Visualization Tools" can be used to examine distributions of gene and pathway BMD values, and to derive a potential point of departure value based on summary statistics. By applying filters on enrichment P-values and minimum number of significant genes, the "Functional Enrichment Analysis" tool enables the user to select biological processes or pathways that are selectively perturbed by chemical exposure and identify the related BMD. The "Multiple Dataset Comparison" tool enables comparison of gene and pathway BMD values across multiple experiments (e.g., across timepoints or tissues). The "BMDL-BMD Range Plotter" tool facilitates the observation of BMD trends across biological processes or pathways. Through our case studies, we demonstrate that BMDExpress Data Viewer is a useful tool to visualize, explore and analyze BMDExpress output files. Visualizing the data in this manner enables rapid assessment of data quality, model fit, doses of peak activity, most sensitive pathway perturbations and other metrics that will be useful in applying toxicogenomics in risk assessment. © 2015 Her Majesty the Queen in Right of Canada. Journal of Applied Toxicology published by John Wiley & Sons, Ltd.

Keywords: 4,4′-methylenebis(N,N-dimethyl)benzenamine (MDMB); BMDExpress; benchmark dose (BMD); bioinformatics; data visualization; dose-response; furan; gene expression; genomics; human health risk assessment; microarray; transcriptomics.

Figure 1

Workflow demonstrating the features and functionality of BMDExpress Data Viewer. BMD, benchmark dose; BMDL, benchmark dose lower confidence (values); GO, gene ontology; POD, point of departure.

Figure 2

Summaries of the analyses performed using the “BMD Analysis Summary” tool. (A) Effects of pre‐filtering methods on the distributions of probe BMD values (furan dataset on the Agilent platform). (B) Distributions of probe BMD values over timepoints (MDMB dataset on the Affymetrix platform). (C) Distributions of fit P‐values (MDMB dataset on the Affymetrix platform). (D) BMD vs. BMDL over time (MDMB dataset on the Affymetrix platform). (E) Distribution of selected models over time (MDMB dataset on the Affymetrix platform). BMD, benchmark dose; BMDL, benchmark dose lower confidence (values); FDR, false discovery rate; MDMB, 4,4′‐methylenebis(N,N‐dimethyl)benzenamine.

Figure 3

Summaries of the analyses performed using the “Functional Classification Summary” tool. (A) Effects of pre‐filtering methods on the distributions of pathway BMD mean values (furan dataset on the Agilent platform). (B) Distributions of pathway BMD mean values over time (MDMB dataset on the Affymetrix platform). (C) Twenty‐four pathways identified by the Dynamic Viewer in the peak regions of the MDMB 90‐day timepoint. Slider bar ranges for BMDL mean, median, BMD fifth and 10th percentile values were selected based on the main modes of their respective histograms. A minimum of five genes was applied as suggested in previous studies (Moffat et al., 2015; Thomas et al., 2011, 2012). BMD, benchmark dose; FDR, false discovery rate; MDMB, 4,4′‐methylenebis(N,N‐dimethyl)benzenamine.

Figure 4

Screenshot of the analysis on the 90‐day timepoint of the MDMB study using the “Functional Enrichment Analysis” tool. Each pathway is represented as a bubble, with size of bubble representing the total number of genes in the biological process, and the depth of color indicating the number of genes from the user's input list that mapped to the pathway. A minimum of five genes and a Fisher's exact test P‐value cutoff of 5% (0.05) were applied in our example analyses. BMD, benchmark dose; MDMB, 4,4′‐methylenebis(N,N‐dimethyl)benzenamine.

Figure 5

Comparison of BMD mean values for the nine common pathways identified across four timepoints in the MDMB study using the “Multiple Dataset Comparison” tool. BMD, benchmark dose; IL, interleukin; MDMB, 4,4′‐methylenebis(N,N‐dimethyl)benzenamine.

Figure 6

BMDL‐BMD range plots for the nine common pathways identified across four timepoints in the MDMB study using the “BMDL‐BMD Range Plotter” tool. BMD, benchmark dose; BMDL, benchmark dose lower confidence (values); MDMB, 4,4′‐methylenebis(N,N‐dimethyl)benzenamine.